How deadly is COVID-19?

6 min readApr 2, 2020

Are the current estimates of COVID-19 fatality rate misleading?

Case Fatality Rate

In epidemiology, the most commonly used metric to measure how deadly a disease or epidemic happens to go by the name of case fatality rate (CFR). By definition, it is a very straightforward concept and it is nothing but the proportion of cases that end up dead during the course of the disease. There are varied definitions of CFR depending on what exactly constitutes the denominator, the number of cases. More often than not, this denominator corresponds to the number of ‘detected cases’, as is in the case of COVID-19. Some find it better to use a different term named infection fatality rate (IFR), where the denominator will include undetected and asymptomatic infections as well as the detected infections. CFR has to be always interpreted as a conditional probability given the constraints or conditions that it was computed on. For example, the current CFR estimates of COVID-19, which are mostly in the range of 2–4%¹ or even more, tells you the risk of dying provided that you get diagnosed with COVID-19 under the current circumstances. It is also to be noted that CFR can be influenced by a lot of factors including quality/capacity of the healthcare environments and demographics of the susceptible population. For example, CFR for older age groups are typically many times more than the CFR for younger age groups for the same infection.

The current COVID-19 fatality rates won’t tell you the bigger picture

It is quite likely that the true fatality rate of COVID-19 is much lesser than the current estimates of 2–4%. In almost all places, the current testing for COVID-19 involves only a subset of symptomatic cases owing to practical reasons. The number of deaths, which is the numerator, is less prone to be biased. Due to this, the current estimates of case fatality rate (CFR) are crude and they are likely to be over-estimated figures because the denominator is not representative of the true number of infections or even the true number of symptomatic cases. The current fatality-rate estimates are still useful when they are strictly inferred as a conditional probability under the constraints that the denominator got constructed, but problems arise when extrapolating the current estimated CFRs to general population or when comparing with related respiratory illnesses like seasonal influenza where CFR estimates (~0.1%) are adjusted for undetected infections². So it doesn’t make much sense to say that COVID-19 is 20–40 times more deadly than seasonal influenza.

We need data from samples that are more representative of general population to get better estimates of fatality-rate

To get better estimates, you would need a sample population that has better potential to generate a more robust denominator (number of infected cases) that is representative of general population.We will naturally have access to better data as we progress, but the question is do we have data from sample populations that are more representative of general population than the current samples?

As of now, there appears to be three sample populations where everyone or a reasonable random sample of the population got tested regardless of the symptoms. These are Iceland, Diamond Princess ship (which got locked down off the coast of Japan in early February) and the Italian municipality of Vo. In the context of COVID-19, these three sample populations are perhaps statistical/epidemiological goldmines. John Ioannidis, one of the most notable clinical epidemiologists of the current era, have briefly spoken about the value of COVID-19 data from the Diamond Princess ship in a rather very controversial article recently³. In Diamond Princess and Vo, literally everyone got tested for COVID-19 and in all the three samples, about half of the infected cases were asymptomatic. The most notable finding is that the fatality rates from these three samples are much lower than almost all the current CFR estimates. The numbers from these populations are shown in the table below.

What do data from these three samples tell us about COVID-19?

Because these three sample populations got tested for COVID-19 as a whole (Diamond Princess, Vo) or randomly (Iceland) regardless of the symptoms, the fatality-rates coming out of these populations have the potential of giving us a more robust view of the actual fatality-rates even though the samples might not appear as completely representative of the general population. About half of the detected cases were asymptomatic and this is quite remarkable as these are people who can still transmit the disease whilst remaining undetected. Putting them into quarantine is thus a potential game-changer and this perhaps contributed significantly towards containment of the epidemic. The Diamond Princess ship had a relatively higher fatality rate (12 out of 712 infected cases died) and the major reason could simply be the fact that this population is much older than others (estimated median age = 64 years). The apparent incidence of COVID-19 in the Diamond Princess was much higher and it shouldn’t be forgotten here that the ship is comparatively a small closed space, with very high transmission potential. Vo in Italy also had relatively more older people, but despite being close to Lombardy, which is probably the worst COVID-19 hit area on the planet, the fatality-rate in Vo is remarkably lower. It is worthwhile to note that Iceland has implemented no lockdown yet although the fact that it is a very sparsely populated nation could be impacting the transmission. It is to be noted that unlike subjects in Diamond Princess and Vo, all of whom have completed almost a month of follow-up since the last reported case, Iceland is still very much alive in terms of the outbreak and this means that we still don’t know about the clinical outcomes of a significant proportion of the infected cases in Iceland. Nonetheless, it is unlikely that the fatality-rate can shoot-up to the levels seen in China, Italy or elsewhere.

UPDATE on 10 April 2020: In the last paragraph, I mentioned about the scenario that we don’t still know the clinical outcomes of all COVID-19 patients in Iceland. When I wrote this article on April 2nd, the number of deaths in Iceland was 2 with translated to a fatality rate of about 0.16%. As of April 10th, there have been 7 deaths in Iceland (5 new deaths in 1 week) and the cases increased to 1675 (wikipedia) from 1220. So, as expected, the fatality rate of Iceland went up to about 0.4% from the earlier 0.16%. This number can still increase as we progress. But like I mentioned before, it is almost certain that the fatality-rate will touch the levels seen in most other countries. Another thing to note is that there doesn’t seem to be any deceleration in terms of COVID-19 testing in Iceland. As of April 10th, they have tested a whopping ~10% of their entire population!

Conclusion

The bottom line is that the actual case/infection fatality rate of COVID-19 is going to be much lesser than the current estimates. The current estimates should not be used to infer conclusions by extrapolating it to the general population. We should also be careful when comparing the currently available COVID-19 fatality-rates with other respiratory viral illnesses like seasonal flu.